completeness

name	completeness
description	Build the all-universe Story Completeness dashboard as an HTML artifact — per-story coverage (bible↔blueprint↔beat), chapter count, blueprint granularity, open gaps, node score, standard deviation, and ballot count, across every universe. Usage /completeness [--fresh]. --fresh re-runs coordination first (no LLM); default reads the latest reports/coordination/*.json.

/completeness — Story Completeness Dashboard

Produce ONE visual dashboard covering every story in every universe: how complete each story is across the three coordinates (MEANING = bible, CONSTRUCTION = blueprint, PROSE = beat), plus its review score, spread, and structure. Read-only. NO votes, NO panels, NO LLM.

Fixed facts

• DB (read-only): sqlcmd -S "(localdb)\MSSQLLocalDB" -d StreetSamurai -Q "<query>" (Windows Auth).
• Story nodes = Nodes.NodeType='story' (TPH discriminator; in EF, db.Nodes.OfType<StoryNode>()).
• Coordination is ss --coordinate --slug <slug> — read-only, no LLM. It writes reports/coordination/<CODE>.coordination.json and stamps ## Beat Coordination Index into the node bible. Run against the built DLL for speed: DLL=$(ls v3/StreetSamurai.Cli/bin/Debug/net*/StreetSamurai.Cli.dll | head -1) then dotnet "$DLL" --coordinate --slug <slug> with export Logging__LogLevel__Default=Warning Logging__LogLevel__Microsoft=Warning to mute EF chatter. Build once if missing: dotnet build v3/StreetSamurai.Cli -c Debug -v q --nologo.
• Score lives on Nodes.Score (trimmed panel figure). SD + ballot count are NOT in NodeScoreHistory (that table is empty) — compute them from the individual persona ballots in NodeReviews, EXCLUDING failed 0-score ballots (draft contamination inflates SD to 30-40).
• Coordination flags: MISSING_MEANING (Beat.Description empty), NO_CONSTRUCTION (beat not mapped to a blueprint slice), NO_PROSE, STUB_PROSE, UNSCORED. UNSCORED is by design — scoring is at the node level (panels), never per beat, so it appears on every story. Report it under the Score columns, not as a coverage gap.
• Invariant (enforced): every story has ≥1 chapter; a single-chapter story prints no "Chapter 1". Show the chapter count so a regression (a flat story) is visible at a glance. Fix a flat story with ss --ensure-chapter --slug <slug> (or --all).

Steps

1. Resolve mode. --fresh → re-run ss --coordinate for every story first (cheap, no LLM), so coverage reflects the current DB. Default → read the existing reports/coordination/<CODE>.coordination.json files (fast). If a story has no report yet, coordinate it regardless.

2. List stories + structure + score. One query gives universe, code, chapter count, direct-beat count (should be 0 — nonzero = flat story regression), and Node.Score:

   SELECT u.Slug AS uni, n.NodeCode AS code, n.Title,
     (SELECT COUNT(*) FROM Nodes c WHERE c.ParentNodeId=n.Id AND c.NodeType='chapter') AS chapters,
     (SELECT COUNT(*) FROM BeatNodes bn WHERE bn.NodeId=n.Id AND bn.IsEnabled=1) AS directBeats,
     CAST(n.Score AS decimal(6,2)) AS score
   FROM Nodes n JOIN Universe u ON n.UniverseId=u.Id
   WHERE n.NodeType='story' ORDER BY u.Slug, n.NodeCode;
   

3. SD + ballots (clean). Compute mean/SD/ballot count from valid ballots only:

   SELECT n.NodeCode,
     COUNT(CASE WHEN r.Score>0 THEN 1 END) AS ballots,
     COUNT(CASE WHEN r.Score=0 THEN 1 END) AS failed,
     CAST(AVG(CASE WHEN r.Score>0 THEN CAST(r.Score AS float) END) AS decimal(6,2)) AS mean,
     CAST(STDEV(CASE WHEN r.Score>0 THEN CAST(r.Score AS float) END) AS decimal(6,2)) AS sd
   FROM Nodes n LEFT JOIN NodeReviews r ON r.NodeId=n.Id
   WHERE n.NodeType='story' GROUP BY n.NodeCode;
   

4. Coverage + gaps per story. From each reports/coordination/<CODE>.coordination.json: totalBeats, covered, storyScope.HasBlueprint, storyScope.Granularity, and flagCounts (surface MISSING_MEANING, NO_CONSTRUCTION, NO_PROSE, STUB_PROSE; fold UNSCORED into the score side). Coverage % = covered / totalBeats.

5. Render ONE artifact. Load the artifact-design skill FIRST, then write an HTML file to the scratchpad dir and publish it with the Artifact tool. Requirements:

   • Grouped by universe; one row per story, sorted by score descending within each universe.
   • Columns: Story (code + title) · Chapters · Beats · Coverage % · Blueprint (granularity) · Open gaps (the non-UNSCORED flags, "—" when none) · Score · SD · Ballots · State.
   • State badge: ✅ aligned (100% covered, blueprint present); ⚠️ gap (coverage <100% or a structural flag); ❌ missing blueprint or flat story (directBeats>0 or chapters=0).
   • Color-code coverage and SD; make the flat-story / missing-blueprint cases impossible to miss.
   • Theme-aware (light + dark), responsive, self-contained (no external assets). Favicon 📊.
   • A one-line legend: "Coverage = beats where MEANING+CONSTRUCTION+PROSE align. UNSCORED is by-design (node-level panels). Every story must have ≥1 chapter."

6. Summarize in chat (below the artifact link): count of aligned vs gapped stories, any flat stories or missing blueprints, and the highest-variance stories (largest SD).

Notes

• Blueprint granularity drives beat↔construction indexing: beat = beat-parallel arrays, chapter = chapter-parallel. Coordination handles both; you only report what it emits.
• To close gaps (not this skill's job — it only reports): ss --backfill-meaning for MISSING_MEANING, ss --generate-blueprint --retrofit for NO_CONSTRUCTION, ss --ensure-chapter for a flat story.